04 December 2008

Published in: Cat risk - ILS

Why catastrophe risk management needs improved exposure data

Hurricanes Katrina, Rita and Wilma showed that insurance brokers and carriers had incomplete and inaccurate data to estimate losses and establish premium fees. What lessons have been learned? By Trish Conway

While the memories of the 2004 and 2005 US hurricane seasons live on, it's relevant to ask how far lessons learned have been incorporated by insurers and reinsurers into their catastrophic risk management practices.

In Louisiana, Mississippi, and Texas, the lessons from Katrina, Rita and Wilma (KRW) relate primarily to improving preparedness and emergency response and rebuilding to stricter standards. This will be essential in mitigating the human and financial impact of future catastrophic storms.

This is no time for complacency. The 2006 and 2007 hurricane seasons passed with no land-falling hurricanes reaching even close to 2005 levels. However, scientists point out that since 1945, there have been only two consecutive years in which there were no hurricane landfalls. Indeed, just a few months ago Gustav and Ike caused estimated insured losses of $20-30 billion, significant, but lower than the $65 billion of insured losses from KRW.

In Europe as in the US, some storm seasons are more active than others. In December 1999 Anatol, Lothar and Martin caused insured losses of over $10 billion. More recently, in 2007, Europe endured the Kyrill wind storm and flooding in the UK in which insured losses reached nearly $11 billion. As with the U.S. market, the question is whether a major catastrophic event would cause severe economic harm to owners and threaten the solvency of insurers. Adding to the risk landscape is the growing concern that global weather patterns may deviate from past experience, under the influence of global warming.

So it is not surprising that catastrophe risk management by insurance companies is becoming a bigger issue than ever. From KRW, insurance companies learned they must work to better understand the technical complexities and challenges of effective catastrophe risk measurement and management, including the quality and completeness of the exposure data they use in catastrophe modeling. The consequences of not taking on this challenge are too important -- including risks to capital levels, earnings, rating agency evaluations and reinsurance costs.

Risk Measurement and Modeling Challenges

In the aftermath of KRW, a variety of serious deficiencies in insurance companies' risk measurement capabilities became evident. It now seems that in some cases commercial carriers underestimated losses - based on existing data, models, and assumptions - by as much as a factor of 10 or more. The failure stemmed from problems with the models themselves and with the accuracy of the property exposure data fed into the models. Even though these issues have begun to be addressed since KRW they have not yet been resolved. For example, a leading modeling company found in a recent study of residential portfolios in Texas that insurance-to-value levels underestimated replacement costs by as much as 40%. Additionally, loss estimates released immediately following Ike's landfall have been revised upwards, and are not yet final.

There are at least three critical areas of vulnerability that insurers must address:

Loss Amplification. Losses from KRW were in many instances far greater than anticipated. A major factor was the significant periods in which business owners and residents could not access their properties because of the destruction of streets and sheer volume of debris left by the storm and flood. Damage to structures was often far greater than first reported and vandalism was widespread. As rebuilding got under way, labour and material availability and prices quickly drove up replacement costs and, inevitably, litigation costs.

Business Interruption. Businesses and their insurers found that they had underestimated the costs of protracted business interruption and of resuming operations. For many businesses, loss of computers and customer data, government slowness in repairing the local infrastructure (roads, power, water, sewage, etc.), and inducing employees to return (and finding living accommodation for them) were more severe than had been estimated.

There were also businesses with unique facilities and operations whose insurance policies did not capture the full costs of repair, replacement and business resumption. For example, numerous coastal resorts, golf courses, and country clubs were severely affected by the storms, but full data about the costs of repairing or replacing these assets was never factored into the underwriting risk assessment. With poor estimated value data used in modeling losses and unrealistically low replacement cost assumptions used in the models, carriers faced huge exposures.

Exposure Data Quality. Arguably, the most dramatic post-KRW revelation from an insurance risk measurement perspective is the incomplete and inaccurate data that insurance brokers and carriers - commercial and residential - relied upon to estimate losses and establish premium fees. Retrospective analysis following the avalanche of KRW claims indicates that incomplete and inaccurate data significantly influenced modeling results, resulting in serious underestimation of values and losses.

The most critical data inadequacies included:

• Undervaluation of structures, contents,and business interruption.
• Absence of or inaccurate data about construction class.
• Absence of year-of-construction data.
• Lack of relevant architectural details that would impact the cost of repairing or replacing structures.
• Incomplete or inappropriate policy information provided by underwriters and/or brokers.

The exposure data quality issue is further exacerbated by carriers' data aggregation practices, which has emerged as a serious issue in both the US and Europe. US companies often aggregate customer exposure data for geographic areas by zip code rather than by longitude and latitude of specific properties. In Europe, aggregation customarily is based on division (e.g., Cresta Zone) and subdivision (e.g., postal codes) within each country.

However, losses from hurricanes can differ markedly between the spot of a coastal landing and even a mile inland or further up or down the coast from the landing. Losses from earthquakes can vary widely within relatively small geographic areas depending on the relative thickness and different composition of soil layers. In Europe, as with Anatol, Lothar and Martin and other notable wind storms, a critical variable in the severity of damage, in addition to the intensity of the storms, is the actual path of each storm - whether it strikes hardest through areas of high population and concentrated economic values or entirely different areas such as offshore oil fields or forests with huge lumber resources.

Improving the Models

For the US, modelers have addressed the need to build in more scenarios about the potential increased frequency and potential different paths of catastrophic weather events. Hurricane risk models now include the potential short-term frequency of storms based on changing weather patterns, not just the long-term frequency based on historical events. The models also now factor in potential increased loss-amplification based on multiple catastrophe-related events and circumstances.

There is also greater recognition that the most common and readily available data about specific sites and neighborhoods - whether submitted by underwriters or obtained from independent data sources - sometimes does not capture all of the relevant attributes and conditions that can impact the severity of loss.

The most significant data-related issue driving model changes is the calculation of insured value of specific properties and structures. Recently revised modelling engines place greater emphasis on detailed property variables. This challenges underwriters to improve the accuracy and completeness of data on the properties they want to insure. The expectations and quality standards on key data are increasing:

• Year of construction - understanding the implications of how older building codes have evolved and obtaining information about the current extent of structure deterioration or recent retrofitting and renovation.
• Construction type - using data to distinguish occupancy-class structures from structures used for non-occupancy purposes.
• Multiple use and multiple occupancy - factoring data about resorts, golf courses, corporate campuses and other properties that contain multiple structures housing varied businesses.
• Location geocoding - differentiating site risk more finitely based on latitude and longitude instead of solely on zip code or even just county. This approach would enable underwriters to look at local infrastructure, accessibility, population density and other factors that could exacerbate business disruption.

The principal result of these refinements and adjustments, modellers say, is that the new models measure the vulnerabilities of structures and commercial operations at specific locations with greater granularity. In turn, this improved quantification invites better decision-making by owners, insurers and reinsurers.

Addressing the Controls around Data Quality

While the models have improved, commercial carriers are becoming more aware that their underwriting processes - particularly property exposure data collection-- lack rigour and sufficient built-in controls. Both independent brokers and underwriters need more training and insight into the purpose and importance of the property data they collect. They must better understand how structural details impact the severity of risk. There are also challenges in obtaining basic data from property owners and brokers - including year and type of construction, square footage, architectural features, and even accurate building addresses. When basic data is finally provided, many carriers still do not look to third party data sources or have internal processes in place to check or corroborate the data for accuracy and validate assumptions.

The quality controls issue also extends to the modellers. They share the challenge of maintaining the completeness and currency of the independent data upon which their models rely and of properly updating the total insured values (TIVs) of the insured properties in their databases. Some models also lack the ability to red-flag potential data inconsistencies or other data-related problems.

Reinsurers' Concerns about Data Reliability

Absorbing the insights from KRW, reinsurers are becoming more aware of both the negative business impact of poor data quality and the lack of granularity of their property risk aggregation practices. They argue that they have no recourse but to add premium risk loads to counter the probability that the underwriters' data collection and aggregation, risk assessment and valuation processes are unreliable. In turn, the insurance industry is responding to such concerns with:

• Underwriting system upgrades. Companies are taking care to ensure that their models are updated and upgraded.
• Stronger internal controls. Companies are becoming more proactive in checking exposure data and output of models. There is more data and scenario testing in which the performance of models of various combinations of data are assessed. They are also test sampling underwriting data to identify aberrant or suspicious entries.
• Improving data reliability. Companies are becoming more disciplined in confirming and corroborating exposure data. They are doing more independent site inspections and using independent data sources - including tax assessment databases, property valuation tools based on the latest replacement costs and GPS imaging sources that provide up-to-date visual data about properties and their locations.
• Re-evaluating exposure aggregation practices. Insurers are working with their commercial clients to reassess their valuation of portfolios based on closer consideration of asset types, values and replacement and business interruption losses.

The Argument for Assuring Exposure Data Quality

As a result, there is a compelling argument that, for more reliable underwriting decision- making, insurers must address the issue of controls around property exposure data quality. One approach would be to use third party assurance services. Across the financial services industry worldwide, the value of independent assurance of the data- collection and management controls companies use in their compliance and reporting is well accepted. Indeed, more than ever the capital markets are looking to such services. Applying assurance discipline and consistency to catastrophe exposure data and management could significantly improve the decision making and pricing of insurers and reinsurers - thus contributing to their financial and competitive success.

Trish Conway is an Actuarial Advisor with Ernst & Young LLP's Insurance and Actuarial Advisory Services. Trish is based in New York City and can be reached at Trish.Conway@ey.com.

Note: The results of the Ernst & Young 2008 catastrophe exposure data quality survey can be found here: Actuarial Raising the bar catastrophe data

Back to top